Land Surface Reflection Differences Observed by Spaceborne Multi-Satellite GNSS-R Systems

Highlights

What are the main findings?

• Distinct spatial patterns: The inter-system reflectivity consistency exhibits a clear spatial distribution. Higher consistency is observed in areas with smoother surfaces and moderate vegetation cover (as characterized by the composite variable VR), while significant discrepancies occur over water bodies.
• Land cover dominance: The reflectivity correlation is strongly influenced by land cover type, with croplands, grasslands, and shrublands showing the highest inter-system consistency.

What are the implications of the main findings?

• Develop targeted correction models: Since surface roughness and vegetation type affect the reflectivity of different systems in distinct ways, corresponding correction models should be established to mitigate systematic biases.
• Prioritize high-consistency regions: Data fusion should account for variations in sensor characteristics and surface properties, with feasibility verification prioritized in cropland and grassland areas where inter-system consistency is relatively high.

Abstract

With the accelerated launch of spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) satellites, GNSS-R has gradually emerged as an important technique for remote sensing. However, due to its pseudo-random observation mode, the use of a single system makes it difficult to provide continuous spatiotemporal coverage over a specific area within the short term. Although interpolation methods can partially alleviate the coverage gaps, their application is limited by accuracy and reliability constraints, which still restrict the practical use of GNSS-R in terrestrial surface monitoring. To address this issue, conducting joint analyses and data fusion of multi-satellite GNSS-R observations has become an important approach to improving the continuity and accuracy of surface monitoring. However, systematic studies on the integration of multi-satellite GNSS-R data remain relatively limited. Moreover, differences in orbital inclination, antenna design, and signal bandwidth among various spaceborne GNSS-R systems lead to discrepancies in their land observations. Therefore, this study systematically analyzes the reflectivity differences among multiple GNSS-R satellites (e.g., the Cyclone Global Navigation Satellite System (CYGNSS), Fengyun-3 (FY-3), and Tianmu-1 (TM-1)) under consistent surface roughness and land cover conditions, with the aim of providing a theoretical and methodological foundation for the fusion and integrated application of multi-satellite GNSS-R data. The results show that, except for desert regions, the spatial distribution of the correlation coefficients from the least squares fitting of reflectivity between different spaceborne GNSS-R satellites exhibits a pattern similar to that of an established variable, i.e., the vegetation–roughness composite variable (VR), with higher inter-system correlations occurring in areas characterized by lower VR values. Significant reflectivity deviations were observed near water bodies and river networks, such as the Amazon, Paraná, Congo, Niger, Nile, Ganges, Mekong, and Yangtze, where both the fitting intercepts and biases are relatively large. In addition, the reflectivity correlations between CYGNSS–TM-1 and CYGNSS–FY-3 are both strongly influenced by surface vegetation cover type. As the correlation increases, the proportion of non-vegetated and forested areas decreases, while that of grasslands, shrublands, and cropland/vegetation mosaics increases. Analysis of inter-system reflectivity correlations across different land cover types indicates that forested areas exhibit low-to-moderate correlations but maintain stable structural characteristics, whereas wooded areas show moderate correlations slightly lower than those of forests. Grasslands, shrublands, and croplands are mainly distributed within regions of moderate surface roughness and correlation, among which croplands have the highest proportion of highly correlated grids, demonstrating the greatest potential for multi-source data fusion. Wetlands display high roughness and low correlation, largely influenced by dynamic water variations, while bare soils show low roughness (0.2–0.4) but still weak correlations.

1. Introduction

The remote sensing mechanism of Global Navigation Satellite System Reflectometry (GNSS-R) enables the retrieval of multidimensional land surface parameters by interpreting the characteristics of GNSS signals reflected from the Earth’s surface [1]. Applications of GNSS-R have been widely demonstrated in ocean monitoring [2,3,4,5], sea ice inversion [6,7,8,9,10,11], soil moisture estimation [12,13,14,15,16,17,18,19,20,21,22], inland water detection [23,24,25,26,27,28], surface deformation monitoring [29], and vegetation biomass retrieval, among other key environmental parameters. Owing to its non-contact sensing nature, wide spatial coverage, and all-weather observation capability, spaceborne GNSS-R has emerged as an important technique for environmental monitoring and land surface parameter retrieval.

The development of GNSS-R satellites has progressed significantly from initial technology demonstration to diverse scientific applications. In September 2003, the United Kingdom Disaster Monitoring Constellation (UK-DMC) satellite, equipped with the first spaceborne GPS reflectometer, was successfully launched, providing the first on-orbit verification of GNSS-R signal reception and demonstrating its application in ocean remote sensing, thereby laying the groundwork for subsequent studies. In July 2014, the United Kingdom launched the TechDemoSat-1 (TDS-1) satellite, which carried the Space Geodesy Radiometer-Remote Sensing Instrument (SGR-ReSI) payload to demonstrate the capability of GNSS-R for accurate global ocean surface wind speed measurements. This mission also explored the potential of GNSS-R for retrieving sea ice extent [10,30], soil moisture [31], sea ice thickness [32], sea surface height [33], and sea ice concentration [34], representing a major advance in applying GNSS-R to both oceanic and terrestrial remote sensing.

In December 2016, as part of NASA’s Earth Venture program, the CYGNSS (Cyclone Global Navigation Satellite System) constellation was successfully launched. Consisting of eight microsatellites in a ~35° inclination orbit, CYGNSS achieved pan-tropical seamless coverage between 38°N and 38°S, with mean and median revisit times of 7 h and 3 h, respectively [35]. Subsequently, in June 2019, China launched the Bufeng-1 (BF-1) mission, which focused on retrieving sea surface wind speeds from GNSS reflections [36]. In December 2019, the Spire Lemur-2 series satellites were launched to receive reflections from GPS, Galileo, and SBAS signals, enabling applications in soil moisture, ocean winds, sea ice, and wetland monitoring [37]. In July 2021, China launched the Fengyun-3E (FY-3) meteorological satellite, where the Global Navigation satellite system Occultation Sounder (GNOS-II) payload integrated GNSS-R and GNSS-RO measurements for retrieving sea surface winds, land surface, and sea ice parameters. In recent years, GNSS-R technology has continued to mature. In January 2023, China successfully launched the Tianmu-1 01 satellite. With the rapid advancement of Global Navigation Satellite Systems (GNSS, e.g., GPS, GLONASS, Galileo, and BDS), Regional Navigation Satellite Systems (RNSS), and Satellite-Based Augmentation Systems (SBAS, e.g., EGNOS, WAAS, and MSAS), the number of satellites in orbit has grown substantially. This expansion provides broader coverage, higher observation density, and continuously improving resolution, resulting in a massive volume of multi-source data. Such data offer a solid foundation for advancing GNSS-R retrievals across a wide range of applications.

With the increasing availability of spaceborne GNSS-R data, a growing number of studies have conducted comparative analyses of land and ocean surface parameters retrieved from different satellite missions. Wang [38] investigated the region spanning 28–33°N and 90–105°E, demonstrating the feasibility of soil moisture retrieval using the Tianmu-1 satellite and highlighting its superior performance compared with CYGNSS. Yang et al. [14] evaluated the capability of the FY-3 GNSS-R constellation to retrieve daily soil moisture at a quasi-global scale and compared its results with those derived from CYGNSS. Similarly, Ma et al. [39] analyzed the sensitivity of FY-3 GNOS-II reflectivity to soil moisture, showing that the sensitivity was consistent across three different GNSS constellations and under varying incidence angles.

Although numerous studies have evaluated the performance of various GNSS-R satellites in land and ocean parameter retrievals, most efforts have concentrated on comparing retrieval outcomes rather than developing data fusion strategies. Research on the integration of multi-source GNSS-R observations to improve retrieval accuracy remains limited. Vegetation cover type and surface roughness are the primary sources of interference in GNSS-R signals, as they alter the emission, reflection, and scattering characteristics of electromagnetic waves at the land surface [21,22,40]. To correct for the effects of surface roughness, researchers have developed a range of physical models and empirical approaches [41]. Meanwhile, vegetation attenuation is commonly quantified using radiative transfer models, such as the τ–ω model [42,43,44]. However, systematic investigations into inter-satellite differences in reflectivity under the combined influence of surface roughness and vegetation cover type are still scarce. Therefore, this study aims to analyze the differences in reflectivity among various satellite systems from the perspectives of surface roughness and vegetation cover type, thereby providing a theoretical and methodological foundation for the future fusion of multi-satellite GNSS-R observations.

2. Materials

This study utilizes data from three pivotal GNSS-R missions: CYGNSS, FY-3, and Tianmu-1 (TM-1). These missions collectively form a core component of the current in-orbit GNSS-R observation infrastructure. CYGNSS offers a mature, science-dedicated dataset often used as a reference. FY-3 embodies the integration of GNSS-R into an operational meteorological service, while TM-1 exemplifies the rapid development of commercial constellations.

2.1. FY-3 Data

The FY-3 series represents China’s second-generation polar-orbiting meteorological satellites. The GNOS-II instrument has been successfully deployed on FY-3E, FY-3F (launched in August 2023), and FY-3G (launched in April 2023), and is planned to be integrated into subsequent FY missions. This will facilitate the development of a more comprehensive multi-GNSS reflectometry satellite constellation system. The relevant parameters of the FY-3 satellite series are summarized in

Table 1. The FY-3 Series of Meteorological Satellites.

Satellite	Orbit Type	Nominal Altitude	Inclination
FY-3E	Near-polar sun-synchronous orbit	836 km	98.75°
FY-3F	Near-polar sun-synchronous orbit	836 km	98.75°
FY-3G	Non-sun-synchronous inclined orbit	407 ± 5 km	50° ± 1°

.

As shown in Figure 1, the specular reflection points of the FY-3 satellite series on 1 January 2024 were mapped. A statistical summary of FY-3 specular reflection observations for the year 2024 indicates that FY-3G, FY-3E, and FY-3F recorded approximately 1.01 × 10⁸, 1.12 × 10⁸, and 1.23 × 10⁸ specular reflection points, respectively, resulting in a total of about 3.36 × 10⁸ observations. On average, ~9.2 × 10⁵ specular reflection points were recorded per day. The contribution of each satellite accounted for 30.2% (FY-3G), 33.2% (FY-3E), and 36.6% (FY-3F), respectively, indicating that FY-3F played a dominant role in terms of observation frequency. This study utilizes FY-3 series data acquired from 1 January to 31 August 2024, with Level-1 product documentation available online for FY-3E (https://img.nsmc.org.cn/PORTAL/NSMC/DATASERVICE/OperatingGuide/FY3E/FY-3E_L1_Data_Instruction_GNOS-II.pdf, accessed on 30 September 2025), FY-3F (https://img.nsmc.org.cn/PORTAL/NSMC/DATASERVICE/OperatingGuide/FY3F/FY-3F_L1_Data_Instruction_GNOS-II_20240523.pdf, accessed on 30 September 2025), and FY-3G (https://img.nsmc.org.cn/PORTAL/NSMC/DATASERVICE/OperatingGuide/FY3G/FY-3G_L1_Data_Instruction_GNOS-II.pdf, accessed on 30 September 2025).

2.2. CYGNSS Data

CYGNSS, supported by NASA and developed by the University of Michigan, is a microsatellite constellation consisting of eight LEO satellites launched in late 2016. It operates in a non-synchronous orbit at an altitude of ~510 km with an inclination of 35°. Each satellite carries a Delay-Doppler Map Inversion (DDMI) with four L1-band channels, enabling the simultaneous reception of four surface-reflected GPS signals, covering the latitude belt between 38°N and 38°S. The Level-1 dataset provides calibrated Delay-Doppler Maps (DDMs) derived from engineering and scientific data for training and evaluation. Generated from L1-band GPS signals (wavelength~19 cm), DDMs are determined by surface scattering geometry, antenna gain, and surface properties, thereby reflecting the dynamic variations in surface moisture. Figure 2 shows the distribution of CYGNSS specular reflection points on 1 January 2024. The study utilizes CYGNSS Level 1 Version 3.2 data from 10 January 2023, to 31 August 2024, with the documentation available at https://cygnss.engin.umich.edu/wp-content/uploads/sites/534/2021/07/148-0137_ATBD-L1B-DDM-Calibration_R3_release.pdf (accessed on 12 November 2025).

2.3. Tianmu-1 Data

TM-1, developed by the China Aerospace Science and Industry Corporation (CASIC) (Beijing, China), is the first commercial meteorological constellation designed to meet the requirements of numerical weather prediction. The constellation was officially put into operation on 9 January 2023. As of October 2025, it has 23 satellites in orbit, capable of simultaneously providing GNSS radio occultation (RO) and GNSS reflectometry (GNSS-R) data. The RO data primarily include Level 1 (L1) products such as additional atmospheric phase information, and Level 2 (L2) products including atmospheric temperature, humidity, pressure profiles, electron density profiles, as well as sea ice and soil moisture. The reflectometry data also consist of L1 and L2 products. TM-1 is capable of simultaneously receiving signals from four global navigation satellite systems—GPS, BDS, Galileo, and GLONASS—enabling global coverage. This study utilizes the TM-1 dataset covering the period from 9 January to 31 December 2023, consisting of L1 reflectometry data from TM-1. The dataset is stored with an hourly sampling frequency. Figure 3 shows the distribution of Tianmu-1 specular reflection points on 10 January 2023.

2.4. Land Cover Data

To investigate the influence of land cover types on the reflectivity response differences among various spaceborne GNSS-R systems, this study employs the MODIS MCD12Q1 IGBP land cover dataset for comparative analysis. MCD12Q1, generated by the MODIS sensor, provides global land cover classification information at a spatial resolution of 500 m. The IGBP classification scheme in MCD12Q1 includes 17 land cover types, as listed in

Table 2. IGBP Classification After Reclassification.

Target Category	Source Category
Forests	Evergreen Needleleaf Forests
	Evergreen Broadleaf Forests
	Deciduous Needleleaf Forests
	Deciduous Broadleaf Forests
	Mixed Forests
Woodlands	Woody Savannas
	Savannas
Grasslands	Grasslands
Shrublands	Closed Shrublands
	Open Shrublands
Cropland/Vegetation mosaic	Croplands
	Cropland/Vegetation mosaic
Permanent Wetlands	Permanent Wetlands
Unvegetated	Urban and Built-up
	Snow and Ice
	Barren
	Water

.

For terrestrial vegetation cover types, in order to clearly analyze the differences among spaceborne GNSS-R satellites across various land cover categories, the MODIS land cover types were reclassified into seven broad classes (Forests, Woodlands, Grasslands, Shrublands, Cropland, Wetlands, and Unvegetated) following the reclassification approach consistent with Friedl et al. [20]. The reclassified land cover distribution is illustrated in Figure 4.

Previous studies have highlighted the influence of incidence angle on reflectivity. For instance, Stilla et al. [45] analyzed the Algeria dunes and Kufrah areas and reported only a minor effect of incidence angle on reflectivity. In this study, spatiotemporally matched data points from CYGNSS, TM-1, and the FY-3 satellite series were extracted, and the reflectivity of ground specular points under different land cover types was compared (Figure 5). The results similarly indicate that the impact of incidence angle on reflectivity differences is relatively small.

2.5. SMAP Surface Roughness Data

This study employs static surface roughness data derived from the soil moisture active passive (SMAP) Enhanced L3 Radiometer Global and Polar Grid Daily 9 km EASE-Grid Soil Moisture (Version 006) product. The dataset provides daily observations from both ascending and descending satellite passes, features a spatial resolution of 9 km by 9 km, and is formatted in HDF5. Measurements from both pass directions were combined to maximize spatial coverage. The data were sourced directly from the European Space Agency website (http://earth.esa.int, accessed on 12 November 2025). For consistency with the SMAP soil moisture data, all other data used in this study were resampled to align with its 9 km EASE-Grid 2.0 projection. Figure 6 presents the static surface roughness data derived from SMAP for the year 2024.

3. Methods

3.1. Calculation of Reflectivity

In GNSS-R land surface sensing, it is commonly assumed that the scattering mechanism is coherently dominated; consequently, surface reflectivity has been established as a conventional and widely adopted geophysical product for comparison.

3.1.1. CYGNSS

Based on the theoretical assumption of coherent reflection over smooth surfaces and incorporating the bistatic radar cross-section (BRCS) values provided by CYGNSS, the surface reflectivity can be calculated using the following formula [17,22]:

$$\Gamma ={\displaystyle \frac{\sigma {(R_t+R_r)}^2}{4\pi {(R_t{R}_r)}^2}}$$

(1)

where $\sigma$ is the CYGNSS BRCS, $R_t$ and $R_r$ denote the distances from the specular point to the transmitter and the receiver, respectively. All the above parameters can be obtained from the official data products provided by the CYGNSS mission.

3.1.2. FY-3 and Tianmu-1

Based on the theoretical assumption of coherent reflection over smooth surfaces and according to the GNSS-R bistatic radar equation [46], the reflectivity observed by the FY-3 satellite is calculated as follows [47]:

$$\Gamma ={\displaystyle \frac{{(R_t+R_r)}^2(P_{DDM}-N)}{4\pi F{(R_t{R}_r)}^2}}$$

(2)

where $P_{DDM}$ denotes the peak power of the DDM, $N$ represents the noise, and $F$ is the BRCS factor of the DDM, which is defined as follows:

$$F={\displaystyle \frac{{\lambda }^2{P}_t{G}_t{G}_r}{{(4\pi )}^3{(R_t{R}_r)}^2}}$$

(3)

where $P_t{G}_t$ represents the GNSS effective isotropic radiated power, $\lambda$ denotes the wavelength of the GNSS signal, and $G_r$ is the receiver antenna gain. All these parameters can be extracted from the FY-3E dataset. All of the aforementioned variables can be obtained directly from the downloaded files.

Regarding the calculation of reflectivity for TM-1 data, the parameter computation follows the same approach as that used for FY-3E. In addition, the TM-1 dataset directly provides reflectivity measurements.

3.2. Data Quality Control

A unified data quality control protocol was applied to CYGNSS, TM-1, and FY-3 series observations:

(1)

Under the same dielectric constant, reflectivity exhibits large variations when the incidence angle approaches 60°. Therefore, GNSS-R observations with incidence angles greater than 60° were excluded to ensure that errors introduced by topographic elevation remain within a reasonable range.

(2)

Observations with receiver antenna gain below zero were removed.

(3)

Measurements with reflectivity values greater than one were discarded.

(4)

Observations with SNR values below zero were excluded. For CYGNSS, the “ddm_snr” was calculated as $10\log (\mathrm{Smax}/\mathrm{Navg})$, whereas for FY-3 and TM-1, “ddm_sp_snr” was computed as $10\lg (\mathrm{S}\text{\_}\mathrm{sp}/\mathrm{N}\text{\_}\mathrm{avg}-1)$. Since this study compares CYGNSS with FY-3 and TM-1, the SNR from FY-3 and TM-1 was converted to match the CYGNSS definition to ensure comparability.

(5)

Data points outside the official SMAP reference range for soil surface roughness (0–1 cm³·cm⁻³) were removed.

(6)

Grid cells classified as water bodies were removed.

3.3. Inter-Satellite Correlation and Error Metrics

For each grid cell, the corresponding time series of CYGNSS data were paired with those of Tianmu-1 and FY-3, respectively. Grid cells with fewer than ten valid temporal samples were considered invalid and excluded from the correlation analysis. The least-squares fitting method was applied to derive the regression parameters between CYGNSS and Tianmu-1, as well as between CYGNSS and the FY-3 series, including slope, intercept, and bias. The slope reflects whether the reflectivity variations between the two systems are consistent, while the intercept represents the systematic bias between them. In addition, the Pearson correlation coefficient between the systems was computed. Specifically, the Pearson correlation coefficient ($R$), bias, and root mean square error ($RMSE$) were considered. $R$ was used to evaluate the spatial variability of reflectivity between systems, bias was employed to quantify systematic deviations, and $RMSE$ was interpreted as the standard deviation of the random error. Their formulations are given as follows:

$$R={\displaystyle \frac{{\displaystyle \sum _{i=1}^n({\Gamma }_i^{TM/FY}-{\overline{\Gamma }}^{TM/FY})({\Gamma }_i^C-{\overline{\Gamma }}^C)}}{\sqrt{{\displaystyle \sum _{i=1}^n{({\Gamma }_i^{TM/FY}-{\overline{\Gamma }}^{TM/FY})}^2}}\sqrt{{\displaystyle \sum _{i=1}^n{({\Gamma }_i^C-{\overline{\Gamma }}^C)}^2}}}}$$

(4)

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{\left({\Gamma }_i^C-{\Gamma }_i^{TM/FY}\right)}^2}}$$

(5)

$$Bias=E\left[{\Gamma }_C\right]-E\left[{\Gamma }_{TM/FY}\right]$$

(6)

where $R$ denotes the correlation between spaceborne reflectivity datasets, ${\Gamma }_i^{TM/FY}$ represents the reflectivity values from Tianmu-1 and FY-3E, and ${\overline{\Gamma }}^{TM/FY}$ is their mean value. Similarly, ${\Gamma }_i^C$ represents the CYGNSS reflectivity, and ${\overline{\Gamma }}^C$ is its mean value. The $RMSE$ between these systematic reflectivity datasets is denoted by $RMSE$. The notation $E$[·] represents the expected value.

4. Results

4.1. Regression Parameters and Spatial Biases Between CYGNSS and Tianmu-1/FY-3

Figure 7 indicates that the spatial distribution of slopes and correlation coefficients obtained from least-squares regression is generally consistent with the VR index proposed by Rahmani [38], which integrates vegetation and surface roughness. Regions with higher inter-system correlations are mostly associated with lower VR values. However, this pattern is not observed in the Sahara Desert, which may be attributed to the highly heterogeneous surface environment, including dunes, flat rock surfaces, mountains, and volcanic terrains.

As illustrated in Figure 7 (CYGNSS vs. TM-1) and Figure 8 (CYGNSS vs. FY-3), both comparisons reveal substantial reflectivity biases near river networks and water bodies. Examples include the Amazon and Paraná rivers in South America, the Congo, Niger, and Nile rivers in Africa, the Ganges River on the Indian subcontinent, the Mekong River in Indochina, and the Yangtze River in China. In these regions, both the bias and the regression intercept are significantly elevated. These discrepancies may arise from: (1) variations in surface roughness and wind conditions that alter signal coherence; (2) differences in orbital altitude and viewing geometry, leading to changes in incidence angle at the specular point; and (3) discrepancies in spatial resolution, whereby differing footprint sizes sample distinct surface areas.

The reflectivity correlation between CYGNSS and TM-1 is relatively high over the Australian Basin, the Indian Peninsula, the Chad Basin in Africa, and the Pampas grasslands in South America. A comparison between CYGNSS and FY-3 shows a similar spatial pattern of correlation variation to that observed between CYGNSS and TM-1.

4.2. Dependence of GNSS-R Reflectivity Correlation on Surface Roughness

As shown in Figure 9, the correlation coefficients between the reflectivity of the CYGNSS–TM-1 and CYGNSS–FY-3 satellite pairs exhibit a similar overall pattern: high correlations are primarily concentrated in regions with surface roughness ranging from 0.2 to 0.8. The results for CYGNSS–FY-3 are particularly evident, displaying a distinct stepped distribution. This indicates that the reflectivity correlation between different GNSS-R systems is dependent on surface roughness.

The dependence of inter-system reflectivity correlations on surface roughness varies across different GNSS-R systems. Conditional probability analysis was conducted, with high correlation defined as r > 0.5 and moderate roughness defined as 0.2 ≤ RC ≤ 0.8. For the CYGNSS–TM-1 pair, the probability of observing high correlation under moderate roughness conditions is 26.10%, whereas it decreases to 16.60% under non-moderate roughness conditions (RC < 0.2 or RC > 0.8).

In contrast, for the CYGNSS–FY-3 pair, the probability of high correlation within the moderate roughness range (0.2–0.8) is 42.19%, compared to 37.57% under non-moderate roughness conditions (<0.2 or >0.8). The relatively small difference between these probabilities suggests that surface roughness exerts a limited influence on reflectivity correlation for this system pair.

Overall, these results indicate that the sensitivity of inter-system reflectivity to surface roughness differs among GNSS-R pairs. Moderate roughness generally favors higher inter-system consistency, although the extent of this influence depends on the specific sensor combination and local surface characteristics.

4.3. Reflectivity Consistency Across GNSS-R Systems Under Different IGBP Land Cover Types

The analysis of reflectivity correlation based on IGBP land cover types indicates that the reflectivity data from both CYGNSS–TM-1 (see Figure 10) and CYGNSS–FY-3 (see Figure 11) exhibit a clear dependence on surface cover. Specifically, as the positive correlation coefficient increases, the proportion of Unvegetated areas and Forests decreases monotonically, whereas the proportion of Grasslands, Shrublands, and Cropland/Vegetation mosaic increases significantly with increasing correlation.

This pattern may be attributed to differences in vegetation characteristics across land cover types. Regions with moderate vegetation cover generally have intermediate biomass, allowing both CYGNSS (L-band) and optical/passive microwave sensors to obtain effective signals. Additionally, the relatively uniform vegetation structure in these areas reduces uncertainty caused by differences in observation angles. Seasonal variations in vegetation growth can also produce synchronized responses in multi-sensor satellite data, enhancing reflectivity consistency across systems.

In contrast, unvegetated areas may be strongly affected by rapid soil moisture fluctuations, to which GNSS-R reflectivity is highly sensitive, leading to lower consistency between systems. Forested areas may experience optical signal saturation due to multi-layer canopy scattering, while microwave signals can penetrate the canopy; this difference in observation mechanisms, combined with complex forest structure, further contributes to inconsistencies between different sensor observations.

4.4. GNSS-R Reflectivity Correlation Across Land Cover and Roughness Gradients

First, the number of valid data points was analyzed. Under conditions of valid correlation coefficients, valid IGBP classifications, and valid soil roughness, the number of valid data points for CYGNSS & FY-3 and CYGNSS & TM-1 were 257,574 and 978,259, respectively. In addition, the proportions of the seven land cover classes for CYGNSS & FY-3 and CYGNSS & TM-1 were calculated, as shown in Figure 12 below.

(1)

Comparison of Correlation and Surface Roughness Features Between CYGNSS and TM-1 Across Different Land Cover Types

The information that can be obtained from Figure 13 is that in forested areas, surface roughness is primarily concentrated in the 0.4–0.8 range, and shows that correlation values are mostly between −0.2 and 0.4, indicating moderate to low levels. Due to canopy shading and complex vegetation structures, significant discrepancies remain between the reflectivity measurements of the two systems. In contrast, in woodland areas, roughness also falls within 0.4–0.8, but correlation values are mainly distributed in the 0.2–0.6 range, showing a certain degree of consistency, although overall stability is lower than in forests.

Grasslands, shrublands, and croplands are primarily distributed in regions of moderate roughness and moderate correlation, with shrublands exhibiting the highest concentration. In high-correlation regions (r > 0.6), the proportion of high-correlation grids in croplands is notably higher than in shrublands and grasslands. This indicates that croplands demonstrate superior inter-system consistency and structural regularity, offering the best potential for multi-source data fusion, whereas grasslands and shrublands are at moderate levels.

Wetlands exhibit roughness mainly in the 0.6–1.0 range, with correlations concentrated around 0.0–0.4. Overall, these areas have higher roughness and lower correlation, heavily influenced by dynamic water bodies and specular reflection interference. Unvegetated (bare) areas show roughness concentrated in the 0.2–0.4 range, but low correlation values indicate limited inter-system consistency.

(2)

Comparison of Correlation and Surface Roughness Features Between CYGNSS and FY-3 Across Different Land Cover Types

According to Figure 14, in forested and woodland areas, surface roughness is mainly concentrated in the 0.2–0.6 range, a range that is slightly lower than the results for CYGNSS–TM-1. This difference may be related to variations in the spatial distribution and number of valid data points. Overall, correlation levels remain generally low. This is primarily due to the high vegetation density in forests and woodlands, where multiple scattering within the canopy reduces the consistency of GNSS-R reflected signals, thereby lowering inter-system correlation.

In contrast, grasslands, shrublands, and croplands exhibit higher overall correlation. Although the roughness distribution is largely similar to CYGNSS–Tianmu-1, the spatial patterns of high-correlation regions differ. This is likely associated with vegetation structure: the short and sparse vegetation in grasslands and shrublands reduces uncertainty in signal scattering paths, while croplands, shaped by tillage activities, display pronounced surface texture orientation, which enhances specular reflection of GNSS-R signals and improves inter-system reflectivity consistency.

Wetlands show scattered distributions of roughness and correlation, without a clear concentration trend. This is mainly attributed to the dynamic distribution of water bodies and the spatial heterogeneity of vegetation, which complicates scattering mechanisms and reduces the stability of correlation.

Unvegetated (bare) areas exhibit roughness mainly in the 0.2–0.4 range, yet inter-system correlation remains low, likely due to significant variations in soil moisture and dielectric properties, which affect reflectivity consistency.

5. Discussion

The consistency analysis of reflectivity measurements from CYGNSS, TM-1, and FY-3 satellites indicates that, under the same observational scenario, the magnitude of reflectivity differences among the spaceborne GNSS-R systems varies with surface characteristics. The spatial distribution of inter-system reflectivity regression parameters, derived from least-squares fitting, exhibits patterns consistent with the VR model proposed by Rahmani et al. [44] across most regions, suggesting that higher consistency among multiple satellite systems tends to occur in areas with smoother surfaces and moderate vegetation coverage. This finding supports previous studies and is likely related to the fact that the coherence of GNSS-R signals is primarily governed by surface scattering conditions and vegetation structure, with variations in surface roughness and dielectric properties serving as key factors controlling reflectivity behavior.

However, this trend does not hold in regions with highly heterogeneous surface environments, such as the Sahara Desert. The diverse geomorphological features in these areas—including sand dunes, rocky plains, and volcanic terrains—may enhance scattering anisotropy and alter the incidence angles at specular reflection points, thereby reducing inter-system correlation. In addition, significant reflectivity biases are observed near major river basins, such as the Amazon, Nile, and Yangtze Rivers, where both the regression intercepts and biases are markedly elevated. The observed variations may also originate from the combined effect of enhanced specular reflection over dynamic water surfaces.

Further analysis of different land cover types reveals that vegetation characteristics play a crucial role in determining inter-system consistency. Grasslands, shrublands, and croplands—representing areas with moderate vegetation cover—exhibit markedly higher correlations than forests and barren regions. This can be attributed to their moderate biomass, relatively homogeneous canopy structures, and stable dielectric properties, which collectively enhance signal coherence between systems. In contrast, forested areas exhibit reduced consistency due to multi-layer canopy scattering and differences in signal penetration depths across wavelengths, while unvegetated (bare) surfaces, despite their low roughness, show weak correlations because of temporal variations in soil moisture and dielectric constants.

Notably, differences between CYGNSS–Tianmu-1 and CYGNSS–FY-3 are also evident. Although both pairs show similar roughness-dependent trends, CYGNSS–FY-3 exhibits a more pronounced stepped distribution of correlation values, likely related to differences in orbital altitude, incidence angle range, and footprint size. This suggests that, in the fusion of multi-satellite GNSS-R datasets, sensor configuration differences and local scattering characteristics must be carefully considered. In practical applications, it is often challenging to disentangle the specific sources of observational errors. These errors can arise from multiple factors: on one hand, from the intrinsic characteristics of the instruments and their calibration; on the other hand, from uncertainties introduced during the retrieval and inversion of geophysical parameters. In addition, gridding the observational data can introduce representativeness errors. It should also be noted that all sensing systems inherently possess some level of error, which constitutes an important source of uncertainty in the observational data.

The relevant parameters of different spaceborne GNSS-R systems vary, including satellite orbital altitude and inclination. As the orbital altitude increases, the projected footprint of the antenna on the ground expands, thereby increasing the number of observable GPS reflection points. However, higher orbital altitudes also lengthen the signal propagation path and reduce the received signal strength, which in turn narrows the usable solid angle of the antenna beam [48]. In addition, differences in the parameters and configurations of various spaceborne GNSS-R sensors should be taken into account. The CYGNSS sensors receive GPS signals, while the GNOS-II instrument onboard FY-3E is capable of receiving both GPS and BDS signals. The GNSS-R payload on Tianmu-1 can receive signals from a wider range of GNSS frequency bands, including GPS L1C/A, BDS B1I/B1C, GAL E1B, GLO L1, and QZSS L1. Future research should focus on developing physically based normalization and correction frameworks that account for incidence angle and roughness effects to further improve multi-source data consistency.

6. Conclusions

This study systematically analyzes the reflectivity consistency of multiple spaceborne GNSS-R missions (CYGNSS, Tianmu-1, FY-3E) across diverse surface roughness and land cover conditions. Even under the same observational scenario, discrepancies exist among these satellite missions and vary according to surface characteristics. The results show that the spatial distribution of regression slopes and correlation coefficients among different systems generally aligns with the distribution pattern of the vegetation–roughness composite variable (VR), with regions of low VR values exhibiting higher inter-system consistency. However, this relationship weakens in highly heterogeneous areas such as deserts and mountainous regions. Significant reflectivity deviations are also observed along major rivers and water bodies, including the Amazon, Nile, and Yangtze Rivers, characterized by increased regression intercepts and reduced consistency, which can be attributed to the dynamic nature of water surfaces. Further analysis reveals that the discrepancies and correlations among the systems vary inconsistently across different scenarios. Moreover, higher consistency is observed over grassland, shrubland, and cropland compared to forest and bare soil, with cropland demonstrating the highest agreement and thus the greatest potential for multi-satellite data fusion. Conversely, forested and wetland areas exhibit lower consistency, which can be attributed to differences in instrument characteristics, multi-layer scattering effects, and the dynamic variations in surface moisture conditions.

Differences in orbital parameters, observation geometry, and footprint size among satellite systems further contribute to the spatial variability in correlation. Future work should focus on developing a unified normalization and calibration framework that accounts for differences in satellite configuration and surface scattering characteristics to enhance inter-system calibration and data fusion accuracy. In summary, this study provides a systematic assessment of reflectivity consistency across multiple GNSS-R systems. The results demonstrate that discrepancies exist in reflectivity measurements from different spaceborne GNSS-R satellites under identical scenarios and that inter-system correlation varies with surface roughness and land cover types, with cropland exhibiting the highest consistency and greatest potential for direct data fusion. Nonetheless, these findings offer theoretical support for multi-satellite GNSS-R data fusion and provide important insights for improving the accuracy of global land surface monitoring.